Directed-Sorting Method for Synthesis of Bead-Based CombinatorialLibraries of Heterogeneous Catalysts
Ramanathan Ramnarayanan, Benny C. Chan, Michael A. Salvitti, andThomas E. Mallouk*
Department of Chemistry, The PennsylVania State UniVersity, UniVersity Park, PennsylVania 16802
Falaah M. Falih, Jason Davis, Douglas B. Galloway, Simon R. Bare, andRichard R. Willis*
UOP LLC, 25 East Algonquin Road, Des Plaines, Illinois 60017
ReceiVed October 9, 2005
The synthesis and analysis of inorganic material combinatorial libraries by a directed-sorting, split-poolbead method was demonstrated. Directed-sorting, split-pool, metal-loaded libraries were synthesized byadsorbing metal salts (H2PtCl6, SnCl2, CuCl2, and NiCl2) and metal standards (Pt, Cu, Ni in HCl) onto 2-mgporousγ-alumina beads in 96- or 384-well plates. A matrix algorithm for the synthesis of bead librariestreated each bead as a member of a row or column of a given matrix. Computer simulations and manualtracking of the sorting process were used to assess library diversity. The bead compositions were analyzedby energy-dispersive X-ray spectroscopy, X-ray fluorescence spectroscopy, electron probe microanalysis,inductively coupled plasma atomic emission spectroscopy, and inductively coupled plasma mass spectroscopy.The metal-loaded beads were analyzed by laser-activated membrane introduction mass spectroscopy(LAMIMS) for catalytic activity using methylcyclohexane dehydrogenation to toluene as a probe reaction.The catalytic activity of individual beads that showed minimal (∼20% of that of Pt on alumina) to highconversion could be determined semiquantitatively by LAMIMS. This method, therefore, provides analternative to screening using microreactors for reactors that employ catalysts in the form of beads. Thedirected-sorting method offers the potential for synthesis of focused libraries of inorganic materials throughrelatively simple benchtop split-pool chemistry.
Introduction
We describe here a split-pool, directed-sorting approachfor the synthesis of inorganic bead libraries. This work buildson an earlier paper1 that demonstrated the use of the split-pool concept in solid-state materials chemistry. The biggestchallenges in our earlier work1 were to develop a taggingscheme to track metal salt adsorption and to avoid the mixingof components and dissolution of the alumina support insequential adsorption steps. These problems resulted inrelatively poor control over bead composition as well as aninability to identify individual beads without postsynthesisanalysis. The directed-sorting approach demonstrated in thispaper, based on matrix methods and adsorption in well plates,enables the synthesis of combinatorial bead libraries withoutproblems of component mixing. The method eliminates theneed for postsynthesis bead identification and also eliminatesthe tagging problem. The sorting algorithm allows one tomake compositionally diverse libraries using inexpensiveequipment (well plates, plastic pipets) in a relatively smallnumber of steps. To demonstrate this approach, we chosenoble metals and 2-mg, porousγ-alumina beads of the kindtypically used in heterogeneous catalysis.2 One library wasevaluated for catalytic activity by laser-activated membraneintroduction mass spectrometry (LAMIMS).3 We chose
methylcyclohexane (MCH) dehydrogenation to toluene as aprobe reaction.
The split-pool method allows one to synthesize a beadlibrary by choosingn components, which in the casedescribed here are adsorbed onto the beads inn well-platesandm split-pool steps. The individual beads are sorted byusing a predetermined algorithm that tracks the history ofeach bead as metal salts are adsorbed onto it. These sortingalgorithms simplify the mapping of a multidimensionalcomposition space into a 2-D array layout. By changing someof the parameters of the sorting algorithm, such as thesequence of row- and column-shuffling steps, it is possibleto change the compositional redundancy of the resultingsplit-pool library. This control is enabled by replacing thevials described in ref 1 by wells in standard commercial wellplates. This simple modification, illustrated in Figure 1,solves the component mixing problem by isolating each beadin a unique well, physically separated from the adjacentbeads. Each bead absorbs the same amount of solution ineach step. The tagging problem is solved by indexing everybead by four coordinates (well plate identity, row and columnnumber, and split pool step). The most important outcomeis the direct correlation of the response of the library (to aprobe reaction or signal) with the composition and sequenceof metal adsorption steps on each bead without physicaltagging.
The concept of spatially addressing a large number ofexperiments can be traced to the work of Mittasch,4 whotested large numbers of catalysts and, in essence, kept trackof every sample and its processing history. Computers werenot invented in 1903, and one can assume that laboratorynotebooks and entries therein served as tags. To ourknowledge, the earliest documented systematic way ofdesigning experiments following a specific algorithm can betraced to the work by Fisher and Yates,5 who were interestedin problems in biology and agriculture, leading to usefulconcepts such as analysis of variance (ANOVA) and designof experiments (DOE). With the advent of combinatorialmethods pioneered by Hanak6 and popularized by numerousresearchers,7-15 sophistication increased along with the useof computers and robotics. Labor-saving DOE constructswere generally not used in this early work, because they werenot easily integrated into the array deposition methods used.16
The method described here utilizes computer simulations ofthe directed-sorting process to gauge the diversity andredundancy of the resulting bead libraries; in principle,screening data from such libraries are amenable to analysisby ANOVA and related statistical methods.
Split-pool combinatorial approaches have been widelyused in organic and bioorganic chemistry and have beenextensively reviewed.2,16,17 Directed-sorting approaches tomaking split-pool materials libraries were reported by ourgroup and by Schunk et al.1,15 Schunk et al.15 describe thesynthesis and characterization of a 3000-member Mo-Bi-Co-Ni-Fe-on-γ-alumina library. Their synthetic procedurefor adsorption involved adsorbing metal salts ontoγ-aluminain a porcelain dish, which was similar to the adsorption invials described in our earlier work.1 Post analysis of beadswas still required in their method to identify bead composi-tions.
Experimental Section
Materials. H2PtCl6, SnCl2‚2H2O, CuCl2‚2H2O, and NiCl2‚6H2O were purchased from Alfa Aesar and used as received.Pt, Co, Cu, and Ni analysis standards (1.0 mg and 10 mg in1.00 mL of 2.00 and 10.0% HCl) were purchased from Hi-
Purity Standards and used as received. Porousγ-aluminabeads made by an oil drop technique18 had a surface area of195 m2/g and an average diameter of 1.5 mm. The averagemass of the support beads was 2.3 mg. Prior to firstadsorption,γ-alumina beads were washed in DI water atroom temperature. The beads were then calcined at 400°Cfor 3 h to remove excess water and allowed to cool toambient temperature. Vaccu-pette 96 (a plastic device fortransferring solutions to standard 96-well plates), well plates,and pipets were purchased from VWR. Cascade Blue,Fluorescein-5-isothiocyanate, Lucifer Yellow, and Sulfo-rhodamine 101 were used as received from Molecular Probes.
One Sphere at a Time (OSAAT) Experiments UnderIncipient Wetness Conditions.In OSAAT experiments, onealumina sphere was placed manually into each well of a 96V-bottom well plate (Nalge Nunc International). To reducethe effects of variable diameter and weight of the beadsduring these experiments, we manually sorted about 700beads by weight (2.3( 0.2 mg) and diameter (sieved usinga 1536-well plate). The beads were then divided into sixequal lots (∼100 beads each). Metal salts were dissolved in0.5 M HCl to a final concentration of 0.09 g/mL. In a typicalprocedure to adsorb 0.05 wt % Pt onto each bead,∼1.8 g ofH2PtCl6 (Alfa Aesar) and 530µL of concentrated hydro-chloric acid were dissolved in 20 mL of water. The solutionswere heated to boiling for 15 min, cooled to room temper-ature, and then diluted to 25 mL. During the adsorptionexperiments, 12µL of solution containing the metal salt wasmanually delivered at room temperature and by means of apipet to each well containing a bead. The beads were thendried at 60°C for 1 h in thewell plates. Adsorption underthese conditions is close to incipient wetness. We alsoexamined the adsorption of Ru, Ni, Au, Pd, Re, Ir, and Rhusing the following metals salts dissolved in 0.5 MHCl: H2PtCl6, RuCl3, RhCl3, Pd(NO3)2, HAuCl4, NiCl2, H2-IrCl6, and NH4ReO4 (Alfa Aesar).
Modifying the Vaccu-pette for Bead Transfer and Useas a Multipipet. In combinatorial chemistry experiments todate and in OSAAT experiments described in our earlierpaper (see refs 1, 2 and references therein), robotic plotters
Figure 1. Schematic illustrating the advantage of replacing beads in vials by beads in well plates to solve the mixing and tagging problems.The wells in each plate are indexed by their rows and columns, and each plate is indexed by a unique identifier that corresponds to onechemical component of the bead library.
200 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
or printers have typically been programmed to deliver themetal salt solutions needed to synthesize large libraries. Here,a Vaccu-pette, which is an inexpensive plastic multitippipettor, was modified to make a multipipet syringe and beadtransfer device. To use the Vaccu-pette as a bead transferdevice, Samco Transfer pipet tips were cut to 4.4-cm lengthand attached to the Vaccu-pette bottom using a 96-well plate.Holes were drilled into the wells of a 96-well plate, and thepipet tips were held onto the Vaccu-pette using the hollowplate and cellophane tape, as shown in Figure 2a and b. Touse the Vaccu-pette as a multipipettor, paraffin wax wasmelted in a crystallization dish. Pipet tips used to deliversolution (0-20 µL) were held onto the pegs of the Vaccu-pette, and the whole assembly was immersed in hot paraffinwax. The wax was cooled to room temperature, and theassembly was removed from the wax using a heated chisel.The bottom of the multipipettor was encased in Parafilm asshown in Figure 2c. This simple set of tools eliminates theneed for expensive synthetic equipment, such as commercialplotters and solid handling devices often used in combina-torial chemistry.
Elemental Analysis.Metal loadings and distribution onrepresentative beads were determined by using energy-
dispersive X-ray spectroscopy (EDS), micro X-ray fluores-cence (µ-XRF), and inductively coupled plasma atomicemission spectrometry (ICP-AES) at UOP and electron probemicroanalysis (EPMA) and inductively coupled plasma massspectrometry (ICP-MS) at Penn State. ICP-AES analysis wasperformed on a Leeman Labs PS3000 using calibrationstandards (Specpure from Alfa Aesar and Hi-Purity Stan-dards). ICP-MS analysis was performed using a FinniganMATELEMENT high-resolution instrument using the samecalibration standards as in ICP-AES. For ICP-MS analysis,the beads were digested for 1 h at 80°C in a mixture of 2mL of DI water, 2 mL of concentrated H3PO4 (EMD fromVWR), and 1 mL of concentrated HCl (EMD from VWR)in Erlenmeyer flasks covered with a watch glass; cooled toroom temperature; and diluted to 50 mL with DI water.Calibration standards containing Cu, Co, and Ni were alsodiluted to 50 mL in the same matrix as the samples. Indium(10 ppb) was used as an internal standard in all samples usedfor ICP-MS, and data were normalized to In counts in themass spectrometer. For some of the samples that had errors>20% by ICP-MS, Cu and Ni calibration standards (Hi-Purity) were spiked together with In, and samples werereanalyzed. EPMA analysis used a Cameca SX-50 bysectioning each bead into two halves and mounting in epoxy.EPMA profiles were recorded from the edge to the centerin 50-µm steps with a 20-µm spot size; because of epoxypenetration, data were not acquired within 20µm of the beadedge. EDS spectra from the bead edge for the beads adsorbedwith Pt were collected on a scanning electron microscope(JEOL 840), which was operated at 15 kV and a beam currentof 600 nA, with counting time of 200 s and spot size of 10µm2. The elemental concentrations were extracted by a linearleast-squares fit to spectra of standards recorded in the sameinstrument under the same conditions. Peak deconvolution,integration, and conversion to wt % were performed usingNoran software. The EDS elemental maps for the beadsadsorbed with Pt, Ru, Ni, Au, Pd, Re, Ir, and Rh werecollected on a scanning electron microscope (JEOL JSM5400) with the IMIX-PC version 10.593. The spectra wererecorded at a takeoff angle of 30°, accelerating voltage of20 KV at a magnification of 75×. Each EDS elemental mapwas collected for 1 h. Under these conditions, the detectionlimits of the EDS system were reached at 0.3 wt %. Wetherefore concluded that the EDS system was able to detectmetal at loadings on the beads of 0.5 wt % or more. Sampleswere also analyzed by micro-XRF (Eagle IIIµProbe fromEDAX) using a Rh-KR source at an excitation energy of25 keV. The spot size was 1 mm, and the collection timeused was 300 s/bead. On the basis of the beam excitationenergy, the escape depth was estimated to correspond to theouter 200µm of theγ-alumina bead.
Directed-Sorting Experiments.The first step in the split-pool synthesis process was adsorption. This was done bytransferring beads using the modified Vaccu-pette shown inFigure 2a. Once metal salt solutions were adsorbed onto thebeads and dried, rows of beads from a given plate weretransferred using the row-sorter shown in Figure 2b to a setof receiving well plates. The process was repeated cyclically.For example, in a four-component split-pool synthesis,
Figure 2. Home-built vacuum pipetting devices for (a) transferringbeads across rows and columns, (b) sorting along rows, and (c)delivering solutions of 0-20 µL.
which involves four receiving well plates, one-fourth of therows are transferred in rows to the first well plate, one-fourthto the next, and so on. This process happens four times perplate for a four-component row shuffle, as illustrated inFigure 3a. The same procedure was used for the columnshuffle, with the transfer being perpendicular to that describedabove. The row- and column-shuffle steps alternated, witha row shuffle after every odd adsorption step (counting thefirst as odd) and a column shuffle after every even adsorptionstep, as shown in Figure 3b,c. These steps are repeated form sorting steps, making the total number of adsorptionsm+ 1. In the sorting algorithm represented by Figure 3 andTables 2 and 3, a shift by one row and one column wasadded after every pair of row/column directed-sorting stepsto maximize the diversity of the bead libraries. For example,in the synthesis of a four-component bead library occupyingfour 96-well plates (12 rows× 8 columns), rows of beadswere moved in groups of three and columns in groups oftwo. In the first row shuffle, rows 1-3 from plate 1 weretransferred to plate 1, rows 4-6 to plate 2, etc. In the secondrow shuffle, rows 12, 1, and 2 from plate 1 were transferredto plate 1, rows 3-5 to plate 2, etc., as illustrated in Figure3c. In the third row sort, rows 11, 12, and 1 from plate 1were transferred to plate 1, rows 2-4 to plate 2, etc. Theadsorption and sorting operations were alternated until thedesired number of split-pool steps was achieved. Thedirected-sorting procedure, thus, replaces vials described inref 1 by well plates.
Test for Methylcyclohexane Dehydrogenation to Tolu-ene Using LAMIMS. The design of the LAMIMS systemused for comparing the activities of catalyst beads has beendescribed previously by Nayar et al.3 Hydrogen gas waspassed through the reactor at 50 mL/min, and a 25-W laser(Synrad) was held over each bead for 6 min at 25% laserpower to reduce metal salts to metal at each bead. The laserwavelength was 10.57-10.63µm, and the spot diameter was3.5 mm. The feed was switched to a mixture of H2 (50 mL/min) and MCH (0.02 mL/min). The laser was switched to55% peak power and held at each bead for 40 s to provideenergy to heat the bead and the gas surrounding the beadand to provide energy needed for the endothermic reaction(MCH to toluene) and for toluene to desorb. Laser energyestimates for bead reduction and LAMIMS analysis are givenin the Supporting Information.
Results and Discussion
Adsorption of Metal Salts onto Individual Beads inWell Plates. We first studied the adsorption of metal saltsonto individual beads to optimize the conditions for synthe-
sizing split-pool bead libraries in well plates. Our initialexperiments involved adsorbing Pt from H2PtCl6 to determinemetal distribution and uniformity onγ-alumina. We devel-oped a one-sphere-at-a-time (OSAAT) approach for theseexperiments. Briefly, one alumina sphere was placed manu-ally into each well of a 96 V-bottom well plate, and one ormore adsorption/drying cycles were performed in that well.In a typical procedure to adsorb Pt onto each bead, beadswere equilibrated in each well plate with 3µL of solution atthe concentration needed for the desired metal weight to beadsorbed onto each bead. Table 1 shows the results of ICP-AES analysis of three beads from each lot, indicating a goodcorrelation between theoretical and measured compositions.The measurements were carried out at six different Pt
a The results shown are for one of four 96 well plates after fourdirected sorting cycles. Each well or bead is identified by a 5-digitadsorption sequence. For example, the first entry in row three(44411) signifies a bead history of three adsorptions of component4 followed by two adsorptions of component 1. The four-digitcomposition below this sequence (3002) indicates that the beadcontains three parts component 4 and two parts component 1. Notethat the last adsorption step (the last digit in the 5-digit sequence)is the same for all beads in a given well plate. Note also that allcompositions sum to 5, which is the number of adsorption steps in4 split-pool cycles.b The algorithm generated all four componentstwice (0005, 0050, 0500, and 5000), 20 binaries with variedredundancy [0041, 0410, 1004, and 4100 (once); 0014, 1400, 0140and 4001 (3 times); 3002 (6 times); 0032, 0320, 2003 and 3200 (5times); 0023, 0230, 0203, 0302, 3020, 2030, and 2300 (6 times)],24 ternaries with varied redundancy [0311, 3110, 1103, and 1031(2 times); 1013, 0131, 1310, and 3101 (3 times); 0113, 1301, 3011,and 1130 (4 times); 1022 and 2210 (10 times); 0122, 2012, 1220,and 2201 (13 times); 0221 (11 times); 2102 (12 times); 0212, 1202,2021, and 2120 (14 times)], and four quarternaries 27 times each(2111, 1211, 1112 and 1121). The missing binaries were 0104,0401, 4010, and 1040.
202 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
Figure 3. Color simulations of the directed-sorting process for a four-component, four-step, split-pool library: (a) first adsorption andone row transfer (first split); (b) second adsorption and first column transfer (second split); (c) third adsorption and second row transfer(third split).
loadings. Analysis indicated that the beads were uniform inPt impregnation ((4% variation from the average). Weconcluded that ICP-AES on Pt was not sensitive to variationin bead weight (7.1%) and bead diameter (7.7%) and possiblyreflected variation in pipetting (3%) (See Supporting Infor-mation for variation in pipetting, bead diameter, and beadweight). Figure 4 shows EDS intensity as a function ofposition on cross-sectioned beads containing Pt. We alsoextended the OSAAT mode of adsorption to Ru, Ni, Au,Pd, Re, Ir, and Rh using 3µL of metal salt solutions in 0.5M HCl and obtained EDS elemental maps on cross-sectionedbeads. We observed shadow artifacts in the EDS mapping,especially with the rhenium and iridium M lines. This artifactmay arise from the low takeoff angle (30°) used for X-raydetection. The effect was less noticeable in the elemental Land M, lines which matched areas where the samples wereflat. Under the conditions of EDS collection, the backgrounddue to the support (γ-alumina) was low. The bright spotson each image (Figure 4) were 3-4 µm in size (based oncalibration of the IMIX software version used in the SEM).We therefore attribute all the bright spots to the presence ofmetal salts and found that they were evenly distributed. Theseexperiments suggested that the OSAAT mode of adsorptioncould be adapted to the synthesis of split-pool libraries usingthe modified Vaccu-pettes as multipipettors and for suctionduring bead transfer.
Adsorption of Metal Salts under Incipient WetnessConditions onto Beads in Well Plate Arrays.The essentialdifference between OSAAT and directed split-pool librarysynthesis was simply the movement of beads betweenadsorption steps. In our original split-pool experiments,1
beads were loaded into 20-mL borosilicate glass vials andthen adsorbed under impregnation (excess solvent) condi-tions. The beads were in close proximity to each other, andwe postulated that eliminating this would greatly reduceheterogeneity and mixing problems. In OSAAT experiments,each bead sat in its own well and adsorbed solution underincipient wetness conditions. Our experiments under OSAATconditions for adsorption of Pt, Ru, Ni, Au, Pd, Re, Ir, andRh produced evenly distributed metal-loaded beads. The useof a well plate addresses the problem of metal ion transferbetween pooled beads. All the beads (in individual wells)could be loaded with metal salt solutions in a single step,dried, and then transferred to the next set of well plates indirected-sorting operations that took the place of pooling and
splitting. We thus optimized adsorption under OSAATconditions and then applied the same conditions to directedcombinatorial bead library synthesis.
Directed-Sorting Algorithm. Bead libraries were syn-thesized using a sorting algorithm conceptually similar tothe randomization of colors in a Rubik’s cube. One of theobjectives in solving the Rubik’s cube puzzle is to start fromdisjoint faces of the cube i.e., a mixture of colors, and endup with uniform color on each face of the cube. This is doneby moving sections of the cube along row or column, similarto viewing each well plate as a matrix (k × l) and shufflingrows and columns. Thus, the 3× 3 Rubik’s cube can bemapped onto a three-layer well-plate structure. The objectiveof our simulations and experiments, however, was thereverse: that is, start with uniform colors (first adsorption)and end up with a disjoint hypercube (the result of manyadsorption steps and sequences in>3 compositional dimen-sions).
Conceptually, one can view each well plate as a beadpositional matrix (k × l) with a third variablej serving asthe well-plate identifier. The number of components in asplit-pool library (n) is fixed by the number of well-plates(indexed byj), analogous to the number of vials or reactionflasks in conventional split-pool synthesis, giving us theflexibility of choosing large or small split-pool libraries.We simulated the transfer sections of each well plate (i.e.,beads along sets of rows or columns) to other well plates,with uniform metal salt adsorption in each well plate betweensorting steps.
To rapidly simulate the sorting algorithms and assess thediversity of bead libraries they produced, a FORTRANprogram (DIRECTSORT) was written, treating beads in wells(that replace vials) as members of a matrix (see SupportingInformation). The program simulated different sorting algo-rithms and ran in a few seconds on a PC, allowing one tochange the algorithm and see the results quickly. The firststep in the process is adsorption of each component onto allthe beads in each well plate. The program treats everyadsorption step as unique in a given well plate and tracksthis by adding the well plate index (j) to every bead (k,l) ofa given well plate.
After the first adsorption there are rectangles of height(h) and width (w), whereh is the total number of rows andw is the total number of columns in every well plate (SeeFigure 3a, top). The next step is to move members of a givenrow (k) in a well plate (j) to the same (j) or another (j + 1,j + 2, etc) plate. This implies that after the first adsorptionand first row transfer there are rectangles of height (h/n) andwidth (w), in each well plate (Figure 3a, bottom). This rowshuffle is followed by a second adsorption step, as shownin Figure 3b (top). At the end of the two adsorptions andone row transfer, each member of the matrix has twocomponents (the first and second adsorption) and is deter-ministically identified. The next step is to move membersof a given column (l) in a well plate to the same or to anotherwell plate. This first column shuffle results in rectangles ofheight (h/n) and width (w/n), as shown in Figure 3b (bottom).The row/column shuffle alternates, with row shuffle afterevery odd adsorption step (counting the first as odd) and
Table 3. Summary Results of Simulations forFour-to-Eight-Component Bead Libraries Made in Standard96-, 384-, and 1536-Well Plates Using the Row-ColumnSorting Algorithm
204 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
column shuffle after every even adsorption step and proceedsto m steps, with total number of adsorptions beingm + 1.The DIRECTSORT program counts the number of uniquesequences of components and compositions (tallied as thetotal number of times a given component is present in theindex of each element) and also gives a map of the finalcompositions.
Figure 3 illustrates the steps in a row/column shuffle usinga four-component, 96-well plate (12 rows and 8 columns)library. The results of four split/pool steps (five adsorptions)are shown in Tables 2 and 3 for this library. This row/columnshuffle algorithm, which also includes a shift of one row orcolumn after each step to increase library diversity (seeExperimental Section) captured 52 of the 56 possiblecompositions. We found that this algorithm missed fourbinaries (combinations of two of four elements) out of apossible 24, accounting for the difference (Table 2). Table3 illustrates the number of unique compositions and se-quences for libraries made with different well-plate sizes andnumbers of split-pool steps.
We can see from Tables 2 and 3 that for small libraries(384 beads), there are as many unique sequences as thereare wells (or beads) in the array, but there is someredundancy of compositions. The simulation procedure tracksindividual beads as they are moved from a row/column of agiven well plate to a row/column of another well plate. Thisprocess, thus, eliminates the need for physical tagging ofbeads. The number of physical manipulations (bead transferand solution adsorption) for ann-component,m-step, directed-sorting library wasm(n2 + 1) + 3n, or on the order ofmn2
(See Supporting Information).Synthesis and Analysis of Metal-Loaded Bead Librar-
ies.For proof-of-concept, a four-component (Pt, Sn, Cu, andNi), four-step (five adsorptions and four transfers), 96-wellplate library was synthesized according to the row/column,directed-sorting algorithm.γ-Alumina beads used in theseexperiments were sorted by a commercial roller grader, and
total metal content after four split pool cycles (five adsorptionsteps) was 0.25 wt % metal. The choice of metals and totalloading was based on previous work on the methylcyclo-hexane (MCH) dehydrogenation reaction by Haensel et al.and Sinfelt et al.19 Analysis of these bead libraries byµ-XRFand ICP-MS gave large relative errors because of the lowloading of metal on the beads.
A second approach (column-only sort) was used to reducethe number of physical manipulations for reasonably diverselibraries and to provide at the same time eight copies of eachbead forµ-XRF and ICP-MS analysis. For simplicity, fouradsorptions of three metal salts was used. The column-onlysort resulted in 3mn compositions and sequences preparedin 4mn steps. The total number of steps can be reduced bya factor of 2 by using suitably designed masks. Thisalgorithm resulted in 36 compositions and used 48 physicalmanipulations for a library of 288 beads. The algorithmproduced four sets of beads containing only one element;nine binary combinations (with two unique orders of additioneach); and three ternary combinations, with one of threecomponents twice (with five unique orders of addition each),as shown in Table 4. We separately synthesized all discretebinary and ternary combinations of elements at the resolutionof four adsorption steps by manual pipetting. In the lattercase, the total number of physical manipulations was 332,four times the number of pipetting steps as beads (82), andfour plate transfers, as shown in Table 5.
To remove discrepancies due to interference from Snand to check for analytical errors up to 1.0 wt %, wesynthesized libraries of Pt, Cu, and Ni using ICP-MSstandards of the three metals in a solution of 2.00% HCl(Hi-Purity Standards) by the column-shuffle algorithm. Wepostulated that using preanalyzed standards would help toreduce analytical errors. Two libraries were synthesized, oneusing the Vaccu-pette (Table 4) and another by manualpipetting (Table 5) using fresh well plates sealed in plasticprior to use.
Figure 4. EDS mapping of composition for a 2-mg bead containing∼0.5% Pt. Each bright spot on the EDS map of the Pt L line variedbetween 3 and 4µm.
The relative errors inµ-XRF, defined as the ratio ofstandard deviation to average of difference in theoreticalamounts and measuredµ-XRF values, were 39% in Pt, 35%in Ni, and 43% in Cu for theoretical metal content of 1.0 wt% on γ-alumina. We concluded that thatµ-XRF was not auseful analytical tool for bead libraries of this type on thebasis of relative errors due to bead diameter (7.7%), beadweight (7.1%), and pipetting (3%).
EPMA Analysis of Metal-Loaded Bead Libraries Made bythe Row/Column Shuffle Algorithm.A few random sampleswere analyzed for Sn content using electron probe micro-analysis (EPMA). Figure 5a and b show elemental profilesas a function of distance along each bead using EPMA,indicating that the metals are distributed throughout eachbead with variation as a function of distance. We alsoobserved that the signal from the EPMA saturated at 0.2 wt% metal.
ICP-MS Analysis of Metal-Loaded Beads Made by theColumn-Only Shuffle Algorithm.The objective of theseexperiments was to test whether using the column shufflealgorithm adsorbed the desired quantities of metal on eachbead, independent of variations in bead diameter, and weight.To compensate for matrix and order of addition effects, wesynthesized mixed solutions of Cu, Co, and Ni from ICP-MS standards of the three metals in 10% HCl (Hi-PurityStandards) by the column-shuffle algorithm to give thecompositions shown in Table 4. We then made a bead libraryby the same algorithm using 10 g of metal (from ICP-MSstandards) per adsorption step and three sorting sequences,resulting in a total metal content of 1.7 wt % after fouradsorptions. One library was synthesized using a Vaccu-pettein fresh well plates that were sealed in plastic prior to use.Pipetting had errors of 5% for all samples, and drifts inplasma resulted in errors of 6%. The instrumental measure-ment errors (confirmed using Cu and Ni spikes in preana-lyzed standards) were 8% in samples containing Cu and Nialone and 15% for both Cu and Ni in combinations of
samples containing Cu and Ni and no Co. The instrumentalmeasurement errors were 10% for samples containing Coalone, 22% for Cu, Ni, and Co (for samples havingcombinations of one or two additions of Ni or Cu or both,and one addition of Co) and 15% for Cu, Ni, and Co (forsamples having two or three additions of Co and the otheradditions from combinations of Cu, Ni, or both). Wepostulated that Co formed cobalt phosphate and experimentedwith addition of spikes of Ni and Cu, ,that is adding morelabile metals, but we were unsuccessful in reducing themagnitude of this analytical error. We also observed thatthe presence of Co led to larger analytical errors in theinstrument.
We examined 22 of 36 compositions in Table 4: onesample each of four adsorptions of each element (1111, 2222,and 3333), one sample each of the 9 binary combinations(2221, 2211, 2111, 3111, 1133, 1333, 2333, 3322, 3222),and 14 samples from ternary combinations (1321, 3211, and2113; 2213, 2132, 3221, and 2321; and 3321, 3132, and3213). The 14 samples of ternary combinations also servedto test order of addition trends in ICPMS analysis. We alsoexamined nine duplicates (2221, 2111, 2213, 2132, 2321,3121, 3111, 1333, and 1223) with and without Ni spikes.The results of predicted versus ICP-MS analytical resultsare part of Table 6. We observed that errors due to ICP-MSwere distributed about two standard deviations around themean for most of the samples. Although the ICP-MSanalytical was not in absolute agreement with the theoreticalamounts, these results confirm the sorting algorithm at thesingle-bead level using ICP-MS. ICP-MS was the onlyanalytical technique we tried that was capable of character-izing individual beads at weight percents below the SEMdetection limit (0.5 wt % for mixtures of metals).
Tests for Catalytic Activity Using LAMIMS. Laser-activated membrane introduction mass spectrometry(LAMIMS) 3 is a rapid screening technique based on mem-brane introduction mass spectrometry (MIMS) developed
Table 4. Adsorption Sequences Made by Column-Only Shuffle for a Three-Component, Three-Step Bead Librarya
1111 2221 2211 3321 2111 3221 3211 1321 3111 1221 1211 23212222 3332 3322 1132 3222 1332 1322 2132 1222 2332 2322 31323333 1113 1133 2213 1333 2113 2133 3213 2333 3113 3133 1213a Each row in the Table represents one of three 96-well plates after three directed-sorting cycles. The order of addition is represented by
the same code as in Table 3. For example, the first entry in column two (2221) signifies a bead history of three adsorptions of component2, followed by two adsorptions of component 1. The algorithm generated beads containing only one element (1111, 2222 and 3333), ninebinary combinations with two unique orders of addition each (2221 and 1222; 2211 and 1221; 2111 and 1211; 3111 and 1113; 1133 and3113; 1333 and 3133; 2333 and 3332; 3322 and 2332; 3222 and 2322) and three ternary combinations with one of three components twiceand with five unique orders of addition each (1132, 1321, 1213, 3211, and 2113; 2213, 1322, 2132, 2321, and 3221; 3321, 1332, 3132,2133, and 3213).
Table 5. Adsorption Sequences from Manual Pipetting in a Three-Component 96-Well Platea
0 0 0 0 0 0 0 0 0 0 0 0a Each well or bead is identified by a four-digit adsorption sequence indicating the order of addition. For example, the first entry in
column two (1112) signifies a bead history of three adsorptions of component 1, followed by one adsorption of component 2. Note that allcompositions sum to 4, which is the number of adsorption steps in 3 split-pool cycles. 0 refers to no adsorption and was used to check forbackground.
206 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
Figure 5. Electron probe microanalysis of metal-loaded beads as a function of position: (a) Pt0.05Ni0.05Cu0.1Sn0.05, (b) Pt0.1Ni0.05Cu0.05-Sn0.05, and (c) Ni0.15Sn0.1. Subscripts indicate metal content in weight %.
earlier by Cooks et al.20 In LAMIMS, the catalyst bead arrayis separated from a mass spectrometer by a siliconemembrane that maintains high vacuum on the spectrometerside but is sufficiently permeable to pass analytical quantitiesof reactant and product gases from the reactor side. Themembers of the bead array are heated serially to the reactiontemperature by a laser. Because the reaction rate is stronglytemperature-dependent, only the heated bead is catalyticallyactive. Product gases detected by the mass spectrometer canthus be assigned to a particular bead, and the entire arraycan be screened by simply steering the laser beam from beadto bead.
We evaluated three well plates (a total of 288 beads)synthesized by the column-shuffle algorithm for catalyticactivity using dehydrogenation of methylcyclohexane (MCH)as a probe reaction. The beads were evaluated in a LAMIMSreactor described in detail in ref 3. In a typical LAMIMSexperiment, 96 metal-loaded beads were placed in the reactor.The metal-loaded beads were reduced under hydrogen bypassing a 25-W laser at 25% of peak power over each beadfor 6 min. The gas flow was switched to a mixture ofmethylcyclohexane (MCH) and hydrogen, the reduced beadswere serially heated, and the mass spectrometer signal atm/z ) 91 was detected as the toluene product signal.3,21 Ata mass-to-charge ratio of 91, the mass spectrometer detectsthe [M - 1]+ fragment of toluene,21 due to loss of a hydrogenatom to form the relatively stable benzyl cation. This cationis thought to undergo rearrangement to form the very stabletropylium cation, and the strong peak atm/z ) 91 is ahallmark of compounds containing a benzyl unit.21
To quantify the mass spectrometer signal, a 1 wt % Pt onγ-alumina catalyst with a metal dispersion of 75%3a servedas an internal standard in each run and was subjected to thesame conditions (reduction and analysis for activity) as thebead samples made by directed sorting. The bead-to-bead
variation in the toluene signal was determined for an arrayof 24 replicate 1 wt % Pt beads, which was subjected tofive serial runs. The average signal varied by 2% over thefive runs, and the relative standard deviation for individualmeasurements was 11%. We calculated an energy balanceto verify that the laser sufficiently heated the samples duringreduction and reaction (see Supporting Information).
We examined 8 replicates of 36 compositions includingpure metals (Pt, Cu, and Ni with 1 wt % onγ-alumina), 9binary combinations (with a redundancy of 2), and 3 uniqueternary combinations (with a redundancy of 5 each). TheLAMIMS area and peak height of the directed-sortingsamples were normalized to the 1 wt % Pt onγ-aluminastandard, described above.
We found that not all samples gave a detectable LAMIMSsignal. The standard deviation in normalized LAMIMSanalysis (area and peak height) was consistent between 0and 0.1. To simplify the analysis, we chose to examine onlycompositions that had a detectable LAMIMS signal (reducingthe number of examined compositions to 23). We also choseaverage LAMIMS peak height and an area of 0.2 as a lowerbound based on the standard deviation in normalizedLAMIMS peak height and area, reducing the number ofcompositions to 16. Seven of the 23 beads that had detectableLAMIMS signals had a LAMIMS peak height and area lowerthan 0.2 (usually on the order of 0.05) with a relative standarddeviation of 50%. Table 7 shows the average, standarddeviation, and relative standard deviation of LAMIMS peakheight and area andµ-XRF analysis of these 16 selectedsamples.
We found that all eight replicate Pt on alumina samplessynthesized by the column shuffle algorithm compared wellto the 1 wt % Pt onγ-alumina standard. The relative errorsin the normalized LAMIMS peak height and area were 9.6%and 11% respectively, indicating that the errors in the
Table 6. ICP-MS Analysis of 22 Bead Samples Made by Directed Sortinga
deviation from theoretical composition and instrumentalstandard deviation (in parentheses), both in percent
a The element code represents the four adsorption steps that resulted in a total of 1.7 wt % metal.
208 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
LAMIMS technique were lower than those of theµ-XRFanalysis (39%) for Pt onγ-alumina. The LAMIMS techniqueusing MCH dehydrogenation to toluene as a probe wassensitive to relative errors in bead weight (7.1%) and beaddiameter (7.7%) and errors due to ICP-AES analysis of 1wt % Pt from Table 2 (4%). We concluded earlier that errorsin the ICP-AES analysis of 1 wt % Pt from Table 2 (4%)were mainly due to pipetting (3%). We estimated a molecule/site ratio of 657 on a one-bead basis, using theoretical Ptsurface area of 276 m2/g,19 and concluded that sites werelimiting under the conditions of the experiment.
We then examined the normalized LAMIMS peak areaas a function of order of addition. Figure 6a-c shows theaverage normalized LAMIMS area as a function of Pt contentand Pt as the last adsorption step (6a) and Ni content andlast adsorption step (6b and c). The error bars in Figure 6a,crepresent the standard deviation in average LAMIMS area.We found only one composition with Cu as the lastadsorption step (PtPtPtCu) had average LAMIMS peakheight and area greater than 0.2, and we included thiscomposition as part of Figure 6a.
On the basis of average normalized LAMIMS area andrelative standard deviation in normalized LAMIMS area anddata from Table 7 and Figure 6a-c, we ranked the 16
selected samples for MCH dehydrogenation (assumingnormalized LAMIMS area measured activity for MCHturnover) as PtPtPtPt> CuPtPtPt> (NiCuNiNidCuPtCuNidPtCuNiNi) > (PtPtCuNidPtNiNiNi) > NiCuCuNi >(CuCuCuNidNiNiNiNi dNiPtCuNi) > (CuCuNiNi ) Pt-CuCuNi) > (PtPtPtCu) NiNiPtPt) > CuNiNiNi.
Overall, the most surprising trend was the variation inactivity with order of addition for the same nominalcomposition in almost all samples that showed activitygreater than 0.2. Figure 6a shows that the activity of Pt-(0.75 wt %)/Cu(0.25 wt %) varied depending on the orderof addition of Pt and Cu to the alumina beads. When Cuwas adsorbed from ICP-MS standard solutions first, followedby three adsorptions of Pt, the average LAMIMS area was47% that of pure Pt. When adsorption of copper followedthree additions of Pt, the average LAMIMS area halved. Weobserved from Figure 6b that the activity of Ni(0.75 wt %)/Cu(0.25 wt %) also varied depending on the order of additionof Ni or Cu to the alumina beads. When the order of additionwas Ni-Cu-Ni-Ni, the LAMIMS area was 44% that ofpure Pt, but when the sequence was Cu-Ni-Ni-Ni, theaverage LAMIMS area halved. Order of addition trends wereseen with a few combinations of Pt, Cu, and Ni consistentlyin a nonpredictive fashion, as shown in Figure 6c.
Table 7. Average, Standard Deviation and Relative Standard Deviation in Normalized LAMIMS Area andµ-XRF of 16Samples Synthesized by the Column-Shuffle Algorithma
a Numbers in parentheses in the first column denote number of samples (out of eight) included in the analysis. The remainder in eachcase had no detectable LAMIMS signal.
The order of addition information could provide a veryvaluable tool in evaluating and optimizing catalysts of thesame nominal composition. We believe that this could bethe most powerful use of the current method. This study alsoevaluated catalytic activity as a function of precise amountsof energy (laser power) and should be useful in evaluatingactivity of catalyst, should availability of energy be anoptimization parameter. As expected, in this model system,we were unsuccessful in finding a combination of elementson alumina that matched the activity of Pt for the MCHdehydrogenation reaction. The next step in combinatorialcatalyst development would be to choose interesting com-positions (based on average LAMIMS peak height and area
comparable to a reference catalyst) and evaluate them inmicroreactors.
Using Direct Sorted LAMIMS Libraries in CatalystDiscovery. The experiments in this paper have quantifiederrors in synthesis and analysis of combinatorial librariesand the subsequent evaluation in a LAMIMS reactor. Thisstudy also showed that very little sample (2 mg) is requiredin analyzing catalytic activity of samples that showedminimal (∼20% of that of Pt on alumina) to high conversion(equivalent to Pt on alumina at 100%) and, therefore,provides an alternative to screening using microreactors22
for reactors that employ catalysts in the form of beads.Gembicki et al.23 argue that catalysts and reactors have
Figure 6. Average normalized LAMIMS area vs theoretical metal content (wt %), showing the effect of order of addition. Error barsrepresent one standard deviation. (a) Three samples with Pt as last adsorption and one with Cu as last adsorption: green closed triangles(NiNiNiNi), blue closed circles (CuPtPtPt), blue open triangles (PtPtPtCu), and black closed diamonds (PtPtPtPt). (b) Samples with Ni aslast adsorption: blue closed triangles (CuCuCuNi), open green triangles (NiCuCuNi), green closed triangles (CuCuNiNi), open red triangles(NiCuNiNi), closed red triangles (CuNiNiNi), and blue closed diamonds (NiNiNiNi). (c) Samples with Ni as last adsorption: red opendiamonds (CuPtCuNi), closed red circles (PtPtCuNi), open blue squares (PtCuCuNi), red asterisk (PtCuNiNi), open black circles (NiPtCuNi),green closed diamonds (PtNiNiNi).
210 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
evolved over the years to operate in a transient mode. TheLAMIMS reactor inherently couples separation and reactionand can be adapted to run as a multifunctional transientreactor. Thus, it is possible that tools such as LAMIMS usedin evaluation of combinatorial libraries will open up newapplications and reactor designs.
Conclusions
The directed-sorting method is a simple benchtop routefor preparing medium to large inorganic material beadlibraries that do not require tagging or postsynthesis analysis.The main advantages of the method are its simplicity, easeof preparation, and low cost. In principle, the use of amultipipettor and corresponding analysis eliminates the needfor robotic plotters that are often used in combinatorialchemistry, resulting in substantial cost savings. ICP-MS wasthe only analytical technique we used that was capable ofanalyzing mixtures of elements at loadings relevant toheterogeneous catalysis at the single bead level. The directed-sorting technique described in this paper also highlights theimportance of order of addition in experiments seeking tooptimize catalytic activity or other catalytic figures of merit.Difficulties with analytical techniques, bead weight, anddiameter variation can by offset in part with the use of achemical probe of bead activity, such as methylcyclohexane,using the LAMIMS reactor configuration.3,24
Acknowledgment. This work was performed with thesupport of the U.S. Department of Commerce, NationalInstitute of Standards and Technology-Advanced Technol-ogy Program (Cooperative Agreement no. 70NANB9H3035)via subcontract with UOP LLC. R.R. thanks Prof. GeorgeAndrews for the references on Fisher and Yates; AnilOroskar and Kurt Vandenbussche of UOP, John Sinfelt ofExxon-Mobil, Prof. Paul B. Weisz and Prof. Chunshan Songat The Pennsylvania State University for stimulating dis-cussions; and John Kittleson and Mark Angelone at ThePennsylvania State University for performing the ICPMS andEPMA analyses.
Supporting Information Available. Experimental andanalytical characterization details, synthetic details, Fortranprogram used to simulate the row/column shuffle algorithm,and energy balance calculations for LAMIMS experiments.This material is available free of charge via the Internet athttp://pubs.acs.org.
References and Notes
(1) (a) Sun, Y.; Chan, B. C.; Ramnarayanan, R.; Leventry, W.M.; Bare, S. R.; Willis, R. R.; Mallouk, T. E.J. Comb. Chem.2002, 4, 569. (b) The work reported in the current paperwas presented at the Symposium on Chemistry of Petroleumand Emerging Technologies (paper PETR-004) at the 230thNational Meeting of the American Chemical Society, Wash-ington, DC, August 2005, and a brief preliminary accountwas accepted for publication in Preprints, American Chemi-cal Society, Division of Petroleum Chemistry:2005, 50 (3),278-281.
(2) (a)Appl. Catal. A2003, 254(1), 1-170, focuses on applyingcombinatorial techniques to heterogeneous catalysis. (b)Macromol. Rapid Commun.2004, 25 (1), 1-386, focuses
on applying tools and instruments used in combinatorialtechniques and describes applications of the combinatorialtechnique to problems in polymers and other materials(semiconductors, catalysts). (c)Meas. Sci. Technol.2005,16 (1), 1-316, is devoted to various aspects of combinatorialmethods.
(3) (a) Nayar, A.; Kim, Y. T.; Rodriguez, J.; Willis, R. R.;Galloway, D. B.; Falaah, H. F.; Smotkin, E. S.Appl. Surf.Sci.2004, 223 (1-3), 118. (b) Nayar, A.; Liu, R.; Allen, R.J.; McCall, M. J.; Willis, R. R.; Smotkin, E. S.Anal. Chem.2002, 74, 1933. (c) Nayar, A.; Liu, R.; Willis, R. R.; Smotkin,E. S. U.S. Patent Application, Publication, 2004/0077096,April 22, 2004. (d) Nayar, A.; Liu, R.; Willis, R. R.; Smotkin,E. S. U.S. Patent 6,923,939, August 2005.
(4) Mittasch, A.AdVances in Catalysis; Academic Press, Inc.:New York, 1950; Vol. 2, pp 81-101.
(5) (a) Fisher R. A.Statistical Methods, Experimental Designand Scientific Inference; Oxford University Press: Oxford,1990. (b)Experimental Design; selected papers of FrankYates; Hafner Publishing Company: Connecticut, 1970.
(8) (a) Moates, F. C.; Somani, M.; Annamalai, J.; Richardson,J. T.; Luss, D.; Willson, R. C.IECR 1996, 35, 4801. (b)Senkan, S.Angew. Chem., Int. Ed.2001, 40, 312. (c)Jandeleit, B.; Schaefer, D. J.; Powers, T. S.; Turner, H. W.;Weinberg, W. H.Angew. Chem., Int. Ed.1999, 38, 2494.
(9) (a) Reddington, E.; Sapienza, A.; Gurau, B.; Viswanathan,R.; Sarangapani, S.; Smotkin, E. S.; Mallouk, T. E.Science1998, 280, 1735. (b) Chen, G.; Delafuente, D. A.; Saranga-pani, S.; Mallouk, T. E.Catal. Today2001, 2443, 1.
(10) (a) Danielson, E.; Golden, J. H.; McFarland, E. W.; Reaves,C. M.; Weinberg, W. H.; Wu, X. D.Nature1997, 389, 944.(b) Danielson, E.; Devenney, M.; Giaquinta, D. M.; Golden,J. H.; Haushalter, R. C.; McFarland, E. W.; Poojary, D. M.;Reaves, C. M.; Weinberg, W. H.; Wu, X. D.Science1998,279, 837.
(11) Van Dover, R. B.; Schneemeyer, L. F.; Fleming, R. M.Nature1998, 392, 162.
(12) (a) Lebl, M.J. Comb. Chem.1999, 1, 3. (b) Lam, K. S.;Lebl, M.; Krchnak, V. Chem. ReV. 1997, 97, 411.
(13) (a) Lam, K. S.; Salmon, S. E.; Hersh, E. M.; Hruby, V. J.;Kazmierski, W. M.; Knapp, R. J.Nature1991, 354, 82. (b)Houghten, R. A.; Pinilla, C.; Blondelle, S. E.; Appel, J. R.;Dooley, C. T.; Cuervo, J. H.Nature 1991, 354, 84. (c)Kozmin, S. A.; Wang, Y.Angew. Chem., Intl. Ed.2003, 42,903.
(14) Miller, C. T.; Mann, G.; Havrilla, G. J.; Wells, C. A.; Warner,B. P.; Baker, R. T.J. Comb. Chem.2003, 5, 245.
(15) Klein, J.; Zech, T.; Newsam J. M.; Schunk, S. A.Appl. Catal.A 2003, 254 (1), 121.
(16) Experimental Design for Combinatorial and High Through-put Materials DeVelopment; Cawse J. N., Ed.; John Wileyand Sons: NJ, 2003.
(17) Combinatorial Chemistry: A Practical Approach; Fenniri,H., Ed.; Oxford University Press: Oxford, UK, 2000.
(18) (a) Mitsche, R. T. U.S. Patent 3943070 19760309, 1976. (b)Steenberg, L. R.; Vesely, K. D. Ger. Offen. Patent DE2048434 19720406, 1972.
(19) (a) Haensel, V.; Donaldson, G. R.IECR1951, 43, 2102. (b)Sinfelt, J. H.; Hurwitz, H.; Schulman, R. A.J. Phys. Chem.1960, 64, 1559. (c) Sinfelt, J. H.J. Mol. Catal. A: Chem.2000, 163, 123 2000. (d) Reyes, S. C.; Sinfelt, J. H.;DeMartin, G. J.J. Phys. Chem B2005, 109, 2421.
(20) (a) Kotiaho, T.; Lauritsen, F. R.; Choudhari, T. K.; Cooks,R. G.; Tsao, G. T.Anal. Chem.1991, 63, 875A. (b) Kotiaho,T.; Lauritsen, F. R.; Choudhari, T. K.; Cooks, R. G.; Tsao,G. T. Anal. Chem.1991, 63, 882A.
(21) NIST Mass Spectra Data Center; Mass Spectra. InNISTChemistry WebBook; NIST Standard Reference DatabaseNumber 69; Linstrom, P. J., Mallard, W. G., Eds.; June 2005,National Institute of Standards and Technology: Gaithers-burg, MD, 20899. http://webbook.nist.gov (accessed June 10,2005).
(22) Baertsch, C. D.; Schmidt, M. A.; Jensen K. F.Chem.Commun.2004, 22, 2610.
(23) Gembicki, S. A.; Vanden Bussche, K. M.; Oroskar, R. A.Chem. Eng. Sci.2003, 58, 549.
(24) (a) Swenson, L. R.; Brandvold, T. A.; McCall, M. J.; Willis,R. R. U.S. Patent 6,844,198, January 2005. (b) Swenson, L.R.; Brandvold, T. A.; McCall, M. J.; Willis, R. R. PCT Int.Appl. WO 2002088674, 2002. (c) Swenson, L. R.; Brand-vold, T. A.; McCall, M. J.; Vanden Bussche, K. M.; Willis,R. R. U.S. Patent 6,808,928, October 2004. (d) Swenson, L.R.; Brandvold, T. A.; McCall, M. J.; Vanden Bussche, K.M.; Willis, R. R. PCT Int. Appl. WO 2002088675, 2002.
CC050137R
212 Journal of Combinatorial Chemistry, 2006, Vol. 8, No. 2 Ramnarayanan et al.
